Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: More parallel search data (results from dual G5)

Author: Robert Hyatt

Date: 21:36:24 05/16/04

Go up one level in this thread


On May 17, 2004 at 00:09:16, Frank E. Oldham wrote:

>On May 16, 2004 at 11:28:40, Robert Hyatt wrote:
>
>>On May 16, 2004 at 10:31:10, Frank Phillips wrote:
>>
>>>A random selection of statistics:
>>>
>>>cpu=4748%  1.88M
>>>cpu=4834%  1.93M
>>>cpu=5094%  3.24M
>>>brettermeier(C)(2466)[73] whispers: ply=12; eval=Mat08; nps=2.86M; time=18.56;
>>>\   cpu=4530%; egtb=0
>>>
>>>Not sure how to interpret the 50 hundred %
>>>
>>>Frank
>>
>>Me either.  That would be good for a 64 CPU machine of course, max=6400% there.
>>But on an 8-way box I assume something in the CPU time is wrong.  In fact, I
>>seem to remember that Sun did something odd.  IE in normal threads the CPU time
>>returned is the time per thread.  But on Solaris (IIRC) CPU is the sum for all
>>threads.  But Crafty does the summing internally.  That would make the above
>>numbers make sense.  Divide by 8 since the CPU time is off by a factor of 8, and
>>the numbers look like 593%, 604%, 637% and so forth...
>
>This is true for Mac OS X also (maybe for any BSD-derived Unix ?) -- crafty
>reports 396% for my dual.
>Frank


I'll see if I can fix it.  Combining all thread times into one return value is
bad, because I would like to be able to time each thread independently.  I do
that so that I can attribute CPU time (thread by thread) to either searching or
SMP overhead.  That is why you usually see my cpu% at 199 vs 200, when I can
guarantee you that normally my dual xeon is only playing chess.  But with
solaris I can't break that out which is a pain...



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.